• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

Assimilation Of Satellite Sea Surface Temperature Retrievals

Posted on: Wednesday, 7 January 2004, 06:00 CST

Data providers and users discuss the key issues for optimal combination of sea surface temperature observations from the wealth of satellite sensors now available.

Sea surface temperatures (SSTs) derived from satellite data, primarily the Advanced Very High Resolution Radiometer (AVHRR) instrument on the National Oceanic and Atmospheric Administration (NOAA) polar orbiters have been available for two decades. The processing methodology has changed little during this period. Now, however, new geostationary sensors retrieve sea surface temperatures with high temporal sampling, resolving the diurnal cycle. In addition, microwave sensors are demonstrating retrieval capabilities- irrespective of cloudiness-that approach the accuracy of current operational clear-sky IR techniques.

The National Environmental Satellite Data Information Service (NESDIS) Office of Research Applications (ORA) convened a workshop at the NOAA Science Center on 24-26 April 2001 to pursue an integrated approach to the optimal assimilation of multisensor satellite data. This required a concerted effort dedicated to the problems of retrieval theory, modeling of surface effects, full characterization of the retrieval error covariances in both space and time, and inclusion of external data such as upper-air analyses of temperature and humidity, wind stress, and surface heat flux. Current assimilation techniques were adapted to make optimal use of the new data types and associated error covariance information. Various assimilation techniques were explored to address issues such as asynoptic observation times.

Thirty scientists attended the workshop. There were presentations from the U.S. Navy, the National Centers for Environmental Prediction (NCEP), Remote Sensing Systems (industry), and universities both national and international.

The Workshop on Assimilation of Satellite SST Retrievals provided a forum for the researchers and the users of the data to present their research and product needs while coming to a rapid consensus on how to proceed on this project.

REQUIREMENTS-CLIMATE, NWP, OPERATIONAL OCEANOGRAPHY, FISHERIES. The main users of the assimilated satellite SST products are the climate community, the operational oceanography community, and fisheries.

Climate. The requirements for climate monitoring are relatively straightforward: get rid of the biases in the data sources. The great (and increasing) volume of satellite observations has the potential to swamp traditional in situ measurements. Additionally, while in situ data are derived from many hundreds of individual sensors, the number of satellites is very much smaller. For these reasons climatologists are most concerned with the biases in the satellite data. Although the current methodologies that blend satellite and in situ data are reasonably effective, it is recognized that there are potential problems, particularly in regions where the latter are sparse, that is, where the remotely sensed data have the greatest impact on the analysis. Problems are particularly acute where sporadic cloud contamination is present, and where corrections are being extrapolated toward the ice edge from the data-sparse Southern Ocean. Thus, better input data (i.e., less initial bias) would make the problem more tractable. There is less concern about the precision of individual retrievals on the time and space scales under consideration. It should be noted that bulk temperature, equivalent to a typical buoy measurement depth, is the desired quantity. The two reasons for this are 1) continuity with the historical record and 2) the bulk temperature represents the store of heat within the ocean mixed layer and is most representative of the integral of the net heat flux into the ocean over time.

Operational oceanography-NWP. The next generation of Numerical Weather Prediction (NWP) models is moving toward a spatial resolution of 10 km globally, and we should be trying to produce an SST analysis at the same scale. Mesoscale forecasting within the next decade will require regional analyses at 1-km resolutions. Surface temperatures of large (and medium) lakes are also of interest, with the Great Lakes being of particular immediate concern. Accuracy requirements are 0.5 K (or better), and coverage should be updated at least daily (preferably every 6-12 h). There are noticeable problems with persistent cloud cover and episodic aerosol contamination, both of which should be ameliorated, if possible. There is interest in SST for tropical cyclone prediction. Also, incorporation of sea ice in the analysis would be welcome.

Navy. The resolution of the finest regional ocean models is now 1/ 32, with operational global runs being performed at 1/8 (soon to be 1/16). This means that there is an immediate need for SSTs at a resolution of about 10 km globally. The primary purpose is to predict currents and thennocline depths, and requires the assimilation of a number of different data types (such as satellite radar altimeter) in addition to SST, as well as fluxes of heat, momentum, and freshwater. As with NWP, the requirements generally concern bias and timeliness/coverage. Ocean models do not respond well to cold anomalies (e.g., cloud or aerosol contamination), which can initiate spurious convection. The primary source of concern at this present time is contamination (cold bias) due to tropospheric aerosols. There is also a desire for high-resolution temperatures in the coastal zone and in the marginal ice zone.

Fisheries. Although no fisheries expert had been specifically invited to the meeting, this industry is one of the main users of satellite SST data. The following is our assessment of the requirements for the industry. There are two main aspects. The first is the pattern of temperature, which is used to locate feeding fish and preferred habitats. For this, clear resolution of fronts, eddies, and upwelling areas is the primary requirement. The second use is for fish stock management, where exact temperatures are needed. These are important for spawning, fish larvae, and identifying conditions of anoxic hypoxia.

ASSIMILATION METHODS. Background. The primary advantage of remote sensing is that it can provide observations in regions for which there are no in situ data available. Current operational methodology involves empirical regression of satellite-measured radiances against collocated in situ data. The retrieval scheme is optimized for the regions where the in situ-satellite matchup data are most abundant, and there is no guarantee that retrievals made in other regions can be made with the same confidence. The assignment of errors to those regions where the in situ data are sparse is currently by implication only. Thus, the error characteristics are best known in regions where satellite data have the least potential impact, and least well known in regions where they have the greatest potential impact.

The methodology of basing retrieval schemes on radiative transfer modeling of satellite radiances and their relationship to surface temperatures and atmospheric conditions were presented and discussed. This approach has matured in recent years, and has been demonstrated to give good quality retrievals in research and operational contexts in Europe (Merchant et al. 1999; Merchant and Harris 1999; Francois et al. 2000; Brisson et al. 1999). Radiative transfer offers a rationale for characterizing and understanding errors on a global scale, not only where in situ observations are available.

Radiative transfer modeling of satellite SST retrievals. The methodology of basing retrieval schemes on radiative transfer modeling of satellite radiances and their relationship to surface temperatures and atmospheric conditions should be pursued. To apply this methodology, the following steps are required: knowledge of error characteristics in all regions, and particularly so in regions with no available in situ data to permit regional bias correction; both the retrieved SST quality and the bias characterization should be validated using all available in situ data; accurate spectroscopy for modeling of satellite radiances is necessary; radiative transfer methods also require accurate characterization of the spectral response and calibration of sensors. Geostationary Operational Environmental Satellite (GOES) radiances should be analyzed for consistency with AVHRR, by comparing observed and predicted distributions of brightness temperatures using radiative transfer models applied to upper-air analyses (e.g., Merchant et al. 1999).

Ancillary data. Use of ancillary data (e.g., NCEP upper-air humidity) may improve retrieval precision and reduce regional and temporal biases. There are three approaches for incorporating such data in the retrieval process. First, the appropriate (precalculated) retrieval coefficient could be chosen in light of the ancillary data. Second, the retrieval could be performed by a neural net that uses both satellite radiances and ancillary data as inputs. Finally, radiance data could be directly incorporated into a physical retrieval algorithm (requiring use of fast radiative transfer; e.g., Nalli and Smith 1998, 2003). High spectral resolution IR sounders/imagers [e.g., Atmospheric Infrared Sounders (AIRS) and Geosynchronous Imaging Fourier Transform Spectro\meter (GIFTS)] are particularly well suited for physical retrievals.

Atmospheric aerosols. Atmospheric aerosol concentrations can significantly exceed background levels during dust outbreaks, biomass burning, and major volcanic events, causing biased retrievals. Modeling of aerosol effects on radiances allowed the large, long-lived biases in ATSR retrievals following the Mt. Pinatubo eruption to be largely eliminated (Merchant et al. 1999; Merchant and Harris 1999). Aerosol modeling should be extended to other sensors. More frequently, dust outbreaks (e.g., from the Saharan region) cause biases over specific regions for several days that have a detrimental impact on operational assimilation. These SST biases can be related to aerosol optical depth observed in channel 1 of AVHRR (available for ~25% of data) and the operational retrievals can be corrected (Nalli and Stowe 2002). The utility of supplementing AVHRR optical depths with forecast fields of aerosol, temperature, and humidity for bias correction should be explored.

Cloud masking. It is essential to mask significant clouds to avoid erroneous cold retrievals. If cold SSTs are assimilated in an ocean model, they can trigger unrealistic downward convection. Current cloud-masking efficiencies are poorly known, as are the retrieval impacts of unmasked subpixel/thin clouds. These impacts should be assessed empirically and by modeling the effects on radiances of different types of cloud contamination and propagating these through to the retrieved SST. This should allow refinement of current cloud masking, including tuning by region and season. The more radical possibility of using ancillary information for improving cloud masking (e.g., water vapor burden, boundary layer and tropopause heights, background microwave SST), and of using geostationary sensor cloud detection to inform polar orbiter data needs to be explored. Cloud masking is traditionally done on a clear/ cloud basis. It should be useful to the assimilation process to indicate the degree of confidence in the clear-sky attribution, so that SSTs that are cooler than the assimilation's first guess can be assimilated with less weight when the cloud tests have been passed marginally.

ROLE OF MICROWAVE SST RETRIEVALS. The combination of microwave (MW) and IR observations will be a major synergism. Microwave SSTs provide the long-term stability lacking in the IR SST. Recent satellite microwave retrievals of SST have been useful for oceanographic research (Wentz et al. 2000; Chelton et al. 2000, 2001). For the first time, oceanographers can view evolving large- scale SST features through clouds on a regular basis. The major advantage of the MW SST is that near-global coverage is obtained in two days by virtue of the fact that microwaves penetrate nonraining clouds with little attenuation. In addition, the MV SSTs are relatively free of large-scale regional biases and time drifts. The major drawbacks to the MW SST are a relatively poor spatial resolution (50 km), increased uncertainty in areas of high winds (> 12 m s^sup -1^), and an inability to make accurate retrievals within about 100 km of coastlines.

IR/MV comparisons. By the end of 2002, there should be three satellite MW radiometers in orbit that can retrieve SST: Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI), Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) on EOS Aqua, and another AMSR on Advanced Earth Observing Satellite-2 (ADEOS-2). These three sensors will provide excellent coverage of the world's oceans. Two of the satellites, TRMM and Aqua, carry both MW radiometers and IR scanners. The TRMM spacecraft offers a unique opportunity to study and compare the MW and IR SST retrievals. Flying with the TMI is the Visible Infrared Scanner (VIRS). The precise collocation of the TMI and VIRS footprints reduces errors caused by rapidly changing phenomena such as fronts, clouds, and rain and allows for a head-to-head comparison. Retrieving SSTs from VIRS by applying the algorithm developed by Schlussel and Albert (2001) should be explored. This study can also be extended to the AMSR and the Moderate Resolution Imaging Spectroradiometer (MODIS) flying on Aqua.

VALIDATION METHODOLOGIES. The conventional approach to validating satellite-derived SSTs is to compare these with coregistered in situ measurements taken by thermometers mounted on buoys and ships. These measurements are taken at a depth of a meter or more and are decoupled from the radiometric surface temperature by vertical temperature gradients that are caused by diurnal heating [up to 4 K; e.g., Webster et al. (1996)] and the thermal skin effect. Spacecraft radiometers detect the signal of the skin temperature, which is generally a few tenths of a degree cooler than the temperature a few millimeters below. The skin-bulk temperature difference is highly variable and can contribute a significant error to the validation exercises. Recent results derived from a validation of AVHRR SSTs against radiometric skin SSTs measured from ships, show a residual uncertainty in the former about half of that determined by earlier comparisons against bulk temperatures (Kearns et al. 2000). Radiometrically determined skin temperatures can therefore be used to validate the satellite-derived SSTs. The Marine-Atmosphere Emitted Radiance Interferometer [M-AERI; Minnett et al. (2001); operates in the range of infrared wavelengths from ~3 to ~18 m and measures spectra with a resolution of ~0.5 cm^sup -1^] is an ideal instrument for validation. M-AERIs have been successfully deployed on many research cruises in a wide range of conditions, and these skin temperature data will be available for validation.

Validation against buoy data. The high-quality in situ validation by the Rosenstiel School of Marine and Atmospheric Science (RSMAS) of satellite retrievals and the collocation and comparison of Polar Operational Environmental Satellites (POES)/GOES and buoy data performed routinely at NESDIS will be used to assess the accuracy of satellite SST retrievals on a global scale.

A validation system designed for the NOAA CoastWatch program will also allow comparisons to be made at high (1-km) spatial resolution for the AVHRR instrument.

Modeling of surface effects. The modeling of surface effects from the oceanic skin layer and diurnal thermocline is necessary because of the diversity of satellite and in situ SST sensors. These satellite and in situ sensors sample SST at different depths and times. To most accurately utilize the unique information from each sensor, these differences must be considered and properly accounted for. Careful modeling of surface effects will be possible to reference the measurements to common depths and times and enable the generation of meaningful comparisons and error statistics.

Skin layer effects are most relevant to the inclusion of infrared satellite sensors. To provide estimates of the bulk temperature with infrared retrievals, one must compensate for temperature differences across the skin layer. Several parameterizations and a growing set of observations can be used to estimate the temperature change. Methods ranging from simple bias adjustments at higher wind speeds to more complicated functions of the net heat flux and wind stress should be evaluated and refined. The formation and evolution of diurnal thermoclines affect measurements from all the sensors. Development and implementation of a new solar flux product using geostationary satellite data was recommended.

The feasibility of deriving surface solar irradiance from satellites on a global scale has already been demonstrated. Irradiances on a global scale were derived from the "historic" radiance data available from the International Satellite Cloud Climatology Project (ISCCP) data (Whitlock et al. 1995). Successful near-real-time retrieval of solar fluxes for the continental United States was demonstrated in Tarpley et al. (1996).

Characterization of observation errors. A critical aspect of current data assimilation is the characterization of observation error. Both IR and microwave SST retrievals are subject to errors that vary seasonally and geographically. The error covariance structures must be accurately quantified on a global basis. Radiative transfer modeling is very useful in estimating such regional and temporal variations in retrieval quality. These estimates can be compared with in situ observations; the highest quality measurements will come from shipborne radiometer observation of skin temperature.

SSTs from infrared satellite data are subject to varying degrees of residual cloud contamination that vary in time and space. The effect of cloud masking deficiencies has been very difficult to characterize, particularly in the absence of in situ data. Comparison of the infrared data to microwave SST retrievals, which are unaffected by clouds, allows quantification of the effect of significant residual cloud contamination, albeit at relatively coarse resolution. Microwave data also allow estimates of cloud liquid water, which is of use in detecting marine stratiform cloud.

A particular advantage of the TMI sensor is that its precessing orbit allows sampling of residual cloud error throughout the diurnal range of conditions. These errors are expected to vary regionally and seasonally. Polar-orbiting sensors tend to be sun synchronous and should be less affected by diurnal variations.

However, a key advantage of geostationary data is that multiple passes can be used to detect cloud motion and hence eliminate pixels that vary significantly from scene to scene. Collocation of geostationary and polar orbiter data will be used to estimate cloud detection efficiency at higher resolutions than are likely to be possible with the microwave/infrared comparisons.

Perhaps the most significant error source in operational assimilation of infrared satellite \SST retrievals is tropospheric aerosol. The aerosol correction methods described above will not be perfect, and the residual errors need to be determined. Microwave SST retrievals can be used to characterize these errors.

An important part of the error characterization is determining the effect of the contamination (expected to be non-Gaussian for several sources). For example, residual cloud contamination almost exclusively results in cold biases for SST retrievals from single- view infrared sensor data and is difficult to correct using a Reynolds-type optimal interpolation (OI) analysis (Reynold and Smith 1994) in regions where in situ data are sparse.

Assimilation of satellite SST retrievals. The U.S. Navy has requirements for real-time, global, high-resolution analyses of SST. These analyses serve multiple and varied purposes at the two agencies (U.S. Navy and NOAA), including the need for ocean nowcasts, ocean prediction, production of synthetic ocean profiles, numerical weather prediction, and coupled air-ocean prediction on a variety of scales from global to local. The Naval Research Laboratory (NRL) proposes to investigate new methods for the optimal assimilation of the diverse SST retrievals into ocean analyses used at the various operational centers.

While today's SST analyses are fairly accurate in areas that are well sampled, there are still problems associated with areas of persistent cloudiness and with the marginal ice zone. Microwave instruments can sample cloudy areas, but the large (50-km) footprint of microwave SST poses a challenge when analyzing these SSTs with much smaller grid spacing, on the order of one to a few kilometers. Typically, the footprint errors associated with assimilation of 4- km AVHRR data have been neglected on 1-km or finer grids. At best, these errors have been categorized as errors of representativeness to reduce the overall confidence in the observed data. The much larger footprint of the microwave data should perhaps be more properly approximated not as a point measurement but as an area average.

Any new data source, even a new method for retrieving a measurement from an existing instrument, must be carefully evaluated. Quality control techniques must be developed and appropriate observation errors must be determined for specification in the SST analyses. In addition, the validity of the final SST analysis products must be evaluated, both by themselves and through their impact on the other operational products.

SUMMARY. The proposed methodology to assimilate satellite sea surface temperature retrievals is the culmination of several maturing technologies to achieve a quantum leap in the operational generation of a key oceanographic, meteorological, and climatological product. It was agreed at the workshop that the methodology proposed here is the correct way to progress.

By treating all sensors within a common framework of radiative transfer modeling, we will be able to integrate polar-orbiting and geostationary IR sensors, with confidence that the integration is globally appropriate. Improvements in SST retrieval methodology will also be developed in connection with aerosol effects cloud contamination and ancillary data. There are important links from this activity to other elements of the proposed methodology. First, there should be an important flow of information and expertise into this activity from the activity on modeling and calculation of diurnal effects (these have different signatures in different sensor SSTs). Second, the modeling of cloud and other errors and biases should feed directly into the overall error characterization. Third, infrared and microwave SSTs should be compared; the infrared part of which should be undertaken within the same radiative transfer framework. Fourth, the SSTs and their error descriptions (along with or in combination with those from microwave sensors) form the major new dataset to be assimilated in ocean models.

REFERENCES

Brisson, A., and Coauthors, 1999: Sea surface temperature retrieval in the framework of the Ocean and Sea Ice Satellite Application Facility: Mid and low latitudes. Proc. Meteorological Satellite Data Users' Conf., Copenhagen, Denmark, EUMETSAT.

Chelton, D., F. J. Wentz, C. L. Gentemann, R. A. DeSzoeke, and M. G. Schlax, 2000: Microwave SST observations of transequatorial tropical instability waves. Geophys. Res. Lett., 27, 1239-1242.

____, and Coauthors, 2001: Observations of coupling between surface wind stress and sea surface temperature in the eastern tropical Pacific. J. Climate, 14, 1479-1489.

Francois, C., and Coauthors, 2000: Sea surface temperature retrieval (NOAA-14 and GOES-8) for the ocean and sea ice. Satellite Application Facility, Meteo-France/CMS Visiting Scientist Report.

Kearns, E. J., J. A. Hanafin, R. H. Evans, P. J. Minnet, and O. B. Brown, 2000: An independent assessment of pathfinder AVHRR sea surface temperature accuracy using the Marine Atmosphere Emitted Radiance Interferometer (MAERI). Bull. Amer. Meteor. Soc., 81, 1525- 1536.

Merchant, C. J., and A. R. Harris, 1999: Toward the elimination of bias in satellite retrievals of skin sea surface temperature. 2. Comparison with in situ measurements. J. Geophys. Res., 104, 23 579- 23 590.

____, ____, M. J. Murray, and A. M. Zavody, 1999: Toward the elimination of bias in satellite retrievals of skin sea surface temperature. 1. Theory, modeling and inter-algorithm comparison. J. Geophys. Res., 104, 23 565-23 578.

Minnett, P. J., R. O. Knuteson, F. A. Best, B. J. Osborne, J. A. Hanafin, and O. B. Brown, 2001: The Marine-Atmosphere Emitted Radiance Interferometer: A high-accuracy, seagoing infrared spectroradiometer. J. Atmos. Oceanic Technol., 18, 994-1013.

Nalli, N. R., and W. L. Smith, 1998: Improved remote sensing of skin temperature using a physical retrieval method. J. Geophys. Res., 103, 10 527-10 542.

____, and L. L. Stowe, 2002: Aerosol correction for remotely sensed sea surface temperatures from the National Oceanic and Atmospheric Administration advanced very high resolution radiometer. J. Geophys. Res., 107, 3172, doi:10.1029/2001JC001162.

____, and W. L. Smith, 2003: Retrieval of ocean and lake surface temperatures from hyperspectral radiance observations. J. Atmos. Oceanic Technol., in press.

Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7, 929-948.

Schluessel, P., and A. Albert, 2001: Latent heat flux at the sea surface retrieved from combined TMI and VIRS measurements of TRMM. Int. J. Remote Sens., 22, 1975-1998.

Tarpley, J. D., and Coauthors, 1996: Experimental GOES short- wave radiation budget for GCIP. Proc. Second Int. Scientific Conf. on the Global Energy and Water Cycle, Washington, DC, National Research Council/ Global Climate Research Programme, 17-21.

Webster, P. J., C. A. Clayson, and J. A. Curry, 1996: Clouds, radiation, and the diurnal cycle of sea surface temperature in the tropical western Pacific. J. Climate, 9, 1712-1730.

Wentz, F. J., C. Gentemann, D. Smith, and D. Chelton, 2000: Satellite measurements of sea-surface temperature through clouds. Science, 288, 847-850.

Whitlock, C. H., and Coauthors, 1995: First global WCRP shortwave radiation budget dataset. Bull. Amer. Meteor. Soc., 76, 1-18.

AFFILIATIONS: HARRIS-Cooperative Institute for Climatic Studies, University of Maryland, College Park, College Park, Maryland; MATURI- NOAA/NESDIS/ORA/Oceanic Research and Applications Division, Camp Springs, Maryland

CORRESPONDING AUTHOR: Eileen M. Maturi, NOAA/NESDIS/ORA/Oceanic Research and Applications Division, 5200 Auth Rd., Camp Springs, MD 20146

E-mail: Eileen.Maturi@noaa.gov

DOI: 10.1175/BAMS-84-11-1575

Copyright American Meteorological Society Nov 2003

More News in this Category